Architecture

Overview

DeepSpeech's architecture consists of:

• Audio Processing Pipeline

• Neural Network Engine

• Language Model

• Inference Engine

• API Layer

Components

• speechd: Speech recognition server

• audioprocessor: Audio preprocessing service

• modelserver: Neural network inference

• KenLM: Language model engine

• VAD: Voice Activity Detection

Scalability Features

• Batch processing

• GPU acceleration

• Distributed inference

• Model parallelism